Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Strategy: Research execution process FMEA — analyzes how the research process itself can fail during execution, distinct from design-level failures.
.claude/skills/yogsoth-ai-process-fmea/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-01 | ✗→✓ | ▲ Improved | -1% | 0% |
| case-02 | ✗→✓ | ▲ Improved | 86% | 0% |
| case-10 | ✗→✓ | ▲ Improved | 24% | 0% |
| case-13 | ✗→✓ | ▲ Improved | -53% | 0% |
| case-14 | ✗→✓ | ▲ Improved | -22% | 0% |
Analyzes failure modes in the research execution process — not what the design gets wrong, but what can go wrong during implementation.
| Parameter | S | M | L | |---|---|---|---| | Process steps analyzed | 4 | 10 | 20 | | Failure modes per step | 2 | 3 | 4 | | Chain depth | 2 | 3 | 5 | | Process controls designed | 2 | 6 | 12 |
function-analysis (process mode) → failure-mode-extraction
→ failure-chain-construction
→ [severity-scoring, occurrence-scoring, detection-scoring] (parallel)
→ action-priority-matrix → mitigation-design-sop
→ re-scoring<!-- BEGIN available-tables (generated) -->
Optional, no fixed order; the final leaf is always a sop.
| Tactic | When to use | | --- | --- | | failure-chain-tracing | Tactic: Trace upstream causes and downstream effects of each failure mode. Builds multi-level cause-mode-effect chains for systemic understanding. | | mitigation-validation | Tactic: Run mini-FMEA on proposed mitigations to verify they do not introduce new failure modes. Prevents mitigation-induced risks. |
Optional, no fixed order; the final leaf is always a sop.
| SOP | When to use | | --- | --- | | action-priority-matrix | Compute Risk Priority Number (RPN = S x O x D), classify failure modes into H/M/L action priority per AIAG-VDA tables. | | detection-scoring | Rate detectability 1-10 (inverted: 10 = hardest to detect). Estimates how likely current controls would catch the failure before impact. | | failure-chain-construction | Build cause-mode-effect chains tracing upstream root causes and downstream cascading effects for each failure mode. | | failure-mode-extraction | Extract structured failure mode list from raw scenarios or artifact analysis. Produces standardized failure mode records. | | function-analysis | FMEA Step 3: Decompose artifact into function tree — identify what each component is supposed to do before analyzing how it can fail. | | mitigation-design-sop | Design prevention, detection, and response measures for high-priority failure modes. Produces actionable countermeasure specifications. | | occurrence-scoring | Rate failure mode occurrence probability 1-10. Estimates how likely each failure mode is to manifest during research execution. | | severity-scoring | Rate failure mode severity 1-10 based on end-effect impact. Follows AIAG-VDA severity scale calibrated for research artifacts. |
<!-- END available-tables (generated) -->
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→pass | 52,180 | 45,839 | -12% | 1 | 1 | 0% | 8,071 | 8,013 | -1% | 0 | 0 | — |
case-06 | pass→pass | 32,711 | 43,863 | +34% | 1 | 1 | 0% | 4,835 | 8,171 | +69% | 0 | 0 | — |
case-02 | fail→pass | 33,362 | 46,828 | +40% | 1 | 1 | 0% | 4,864 | 9,068 | +86% | 0 | 0 | — |
case-03 | pass→pass | 22,066 | 11,855 | -46% | 1 | 1 | 0% | 1,465 | 1,712 | +17% | 0 | 0 | — |
case-04 | pass→pass | 8,260 | 13,230 | +60% | 1 | 1 | 0% | 411 | 2,061 | +401% | 0 | 0 | — |
case-05 | pass→pass | 18,734 | 31,262 | +67% | 1 | 1 | 0% | 2,125 | 5,390 | +154% | 0 | 0 | — |
case-07 | pass→pass | 20,595 | 52,636 | +156% | 1 | 1 | 0% | 2,761 | 9,044 | +228% | 0 | 0 | — |
case-08 | pass→pass | 26,322 | 46,670 | +77% | 1 | 1 | 0% | 4,191 | 9,034 | +116% | 0 | 0 | — |
case-09 | pass→pass | 16,380 | 21,831 | +33% | 1 | 1 | 0% | 1,951 | 4,630 | +137% | 0 | 0 | — |
case-10 | fail→pass | 22,719 | 30,793 | +36% | 1 | 1 | 0% | 3,774 | 4,695 | +24% | 0 | 0 | — |
case-11 | pass→pass | 28,423 | 46,671 | +64% | 1 | 1 | 0% | 3,354 | 7,908 | +136% | 0 | 0 | — |
case-12 | pass→pass | 14,355 | 8,994 | -37% | 1 | 1 | 0% | 2,398 | 1,532 | -36% | 0 | 0 | — |
case-13 | fail→pass | 21,937 | 8,368 | -62% | 1 | 1 | 0% | 3,024 | 1,420 | -53% | 0 | 0 | — |
case-14 | fail→pass | 48,146 | 2,523 | -95% | 1 | 1 | 0% | 1,601 | 1,255 | -22% | 0 | 0 | — |
case-23 | pass→pass | 21,272 | 16,046 | -25% | 1 | 1 | 0% | 3,737 | 3,220 | -14% | 0 | 0 | — |
case-15 | fail→pass | 18,643 | 5,576 | -70% | 1 | 1 | 0% | 1,901 | 1,686 | -11% | 0 | 0 | — |
case-16 | pass→pass | 9,645 | 19,592 | +103% | 1 | 1 | 0% | 1,442 | 1,857 | +29% | 0 | 0 | — |
case-17 | pass→pass | 16,488 | 23,658 | +43% | 1 | 1 | 0% | 2,477 | 3,700 | +49% | 0 | 0 | — |
case-18 | fail→pass | 18,275 | 16,494 | -10% | 1 | 1 | 0% | 2,758 | 3,500 | +27% | 0 | 0 | — |
case-19 | pass→pass | 9,528 | 10,084 | +6% | 1 | 1 | 0% | 1,404 | 2,378 | +69% | 0 | 0 | — |
case-20 | pass→pass | 16,028 | 8,401 | -48% | 1 | 1 | 0% | 1,730 | 1,963 | +13% | 0 | 0 | — |
case-21 | fail→fail | 12,288 | 8,038 | -35% | 1 | 1 | 0% | 1,538 | 1,355 | -12% | 0 | 0 | — |
case-22 | pass→pass | 19,138 | 18,691 | -2% | 1 | 1 | 0% | 2,548 | 3,365 | +32% | 0 | 0 | — |
DecimalAI ran this skill against gemini-3.6-flash twice over the same eval suite — once with the skill loaded and once without — and compared the two runs case by case. 23 cases were attempted, and 22 counted toward the lift figure. The other 1 produced results that are not comparable between the two arms, so they are excluded from the headline rather than averaged into it. The headline lift of +30 percentage points is the difference between those two pass rates over the 22 comparable cases.
Without the skill loaded, the model failed this case. With it loaded, the same prompt on the same model passed. This is one improved case from the latest verified run; every case, including any that regressed, is in the table above.
Other measured skills in the registry, with their headline benchmark lift.